Revolutionizing Fertilizer Analysis: Raman Spectroscopy and Machine Learning Deliver Precision and Speed

Farmers rely on fertilizer to help grow their crops and produce the greatest harvest yield. As a result, they have an interest in knowing the nutrient content in the fertilizer that they use, so they can ensure they are using the most optimal fertilizer for their crops.

A recent study led by researchers from Kunming University of Science and Technology recently explored this topic. In a study published in Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, the researchers proposed a new efficient method to detect and analyze fertilizer nutrients using Raman spectroscopy combined with machine learning (ML) (1). Their study demonstrates that their method could potentially address several key problems in traditional fertilizer analysis, including the difficulty in quantifying multiple components and risks of cross-contamination (1).

Feeding lawn with granular fertilizer for perfect green grass | Image Credit: © ronstik - stock.adobe.com

In essence, fertilizer is plant nutrition (2). Without fertilizer, plants cannot grow to their full potential, which is why farmers rely on them (2). However, bad fertilizer, or overusing it, can be counterproductive. Using too much fertilizer can ultimately kill the plant or crop (3). Therefore, the rapid and accurate detection of fertilizer nutrient information is critical to modern agricultural practices. Traditional techniques, while effective, are often time-consuming and limited in their ability to analyze multiple components simultaneously (1). The research team attempted to resolve these challenges by using Raman spectroscopy and integrate it with ML algorithms (1).

Their study focused on five commonly used fertilizers: potassium sulfate (K₂SO₄); urea (CO(NH₂)₂); monopotassium phosphate (KH₂PO₄); potassium nitrate (KNO₃); and a composite fertilizer (N:P:K 15-15-15). Using five spectral preprocessing methods, including Savitzky-Golay (S-G) smoothing, standard normal variate (SNV), multiplicative scatter correction (MSC), detrending (DT), and normalization (NOR), these preprocessing methods were designed to improve the quality of the Raman spectra (1). These preprocessing techniques were crucial in improving the accuracy and robustness of the calibration models.

The researchers then tested three classification algorithms to see how effective they were at identifying the five fertilizer types. These classification algorithms were random forest (RF), particle swarm optimization support vector machine (PSO-SVM), and radial basis function neural network (RBFNN). Out of these algorithms, RF achieved the best classification accuracy at 100%, demonstrating its potential utility in fertilizer identification (1).

For the quantitative analysis part of the study, the researchers compared partial least squares regression (PLSR), extreme learning machine (ELM), and least squares support vector machine (LSSVM). Out of these, PLSR emerged as the top-performing algorithm, with prediction coefficients of determination (R_p²) ranging from 0.9843 to 0.9990 and root mean square errors (RMSE) between 0.0486 and 0.1691, indicating great accuracy in predicting fertilizer concentrations (1).

An interesting wrinkle to this study was the researchers’ examining different water types and how they impact the detection of fertilizer nutrients. Although the water type, whether it was deionized water, well water, or industrial transition water, did not significantly affect the qualitative identification of fertilizer components, it had a noticeable influence on the quantification of nutrient concentrations (1). This finding showcases the importance of accounting for water composition in real-world applications of Raman spectroscopy (1).

As the researchers showed in their study, integrating Raman spectroscopy with ML can help farmers and researchers rapidly, non-destructively, and accurately detect multicomponent fertilizer solutions (1). By optimizing spectral preprocessing and leveraging advanced machine learning algorithms, the researchers have developed a robust framework that ensures high accuracy and reproducibility in both qualitative and quantitative analyses (1).

Modern agricultural practices are concentrated on reducing waste and enhancing crop yield. Using Raman spectroscopy has become one way to accomplish both these objectives. This study is further evidence that further exploration into how Raman spectroscopy can be used to detect other agricultural inputs and contaminants (1).

Having the right fertilizer and dispensing it in the proper amount is key to ensure agricultural efficiency and success. Farmers are constantly faced with the important decision of selecting the right fertilizer. Choosing the right one takes more than just knowing the right balance of nitrogen, phosphorus, and potassium content (3). Thanks to Raman spectroscopy and ML algorithms, it has become easier for farmers to select the right fertilizer for their plants.

References

1. Li, J.; Ma, Y.; Zhang, J.; Kong, D. Rapid Detection of Fertilizer Information based on Raman Spectroscopy and Machine Learning. Spectrochimica Acta Part A: Mol. Biomol. Spectrosc. 2025, 324, 124985. DOI: 10.1016/j.saa.2024.124985
2. Agricultural Industries Confederation, Why We Need Fertilizers. Agricultural Industries Confederation. Available at: https://www.agindustries.org.uk/resource/why-we-need-fertilisers.html (accessed 2025-01-07).
3. University of Minnesota Extension. Ask Extension: Do Fertilizers Help or Hurt Plants? UMN.edu. Available at: https://extension.umn.edu/yard-and-garden-news/ask-extension-do-fertilizers-help-or-hurt-plants (accessed 2025-01-07).